Create and manage Doozies

A Doozy is a persistent AI teammate inside one workspace. It has its own job description, memories, attached files, skills, responsibilities, working hours, toolkit, and approval behavior. It can work independently or coordinate with other Doozies.

The role persists across work contexts. You can open several conversations or to-dos with the same Doozy in separate tabs (aka task contexts, work contexts, or sessions), so a long-running task doesn’t block a new one or mix the two histories. Create another Doozy when the work needs a different role, memory, toolkit, or approval boundary.

Give each Doozy a clear job

Create roles around durable ownership rather than a single temporary to-do (aka task or work item). Customer Researcher, Release Engineer, and Executive Operations create clearer expectations than General Helper.

A useful job description explains:

• what the Doozy owns;
• what sources it should trust;
• how complete work should look;
• which decisions it can make;
• when it must ask for approval;
• which other Doozies it should involve.

You can write this in normal prose. Doozy will work with incomplete descriptions and learn through corrections, but explicit boundaries matter for consequential actions.

Create a Doozy

Open Doozies in the workspace and choose New Doozy. Add a name and describe the job. After creation, you can attach reference files and configure the rest of the role.

During workspace onboarding, Doozy can also suggest and create a small team based on your goals and connected context. Review the proposed roles before accepting them; you can change everything later.

Configure the role

Each Doozy can have:

• Job description: its overall purpose, operating guidance, and boundaries.
• Responsibilities (aka standing ownership or recurring duties): recurring outcomes it owns and keeps visible in the to-do system.
• Skills: reusable procedures, examples, and attached reference files it can load when needed.
• Connections: the workspace apps and toolkits available to this Doozy.
• Files: stable role context such as policies, rubrics, templates, or product references.
• Working hours (aka availability window): when proactive and scheduled work should normally occur.
• Approvals: instructions that classify actions requiring human review.

Keep the role readable. Put lengthy, to-do-specific procedures in skills (aka playbooks, procedures, or prompt templates) rather than turning the job description into a manual for every possible kind of work.

Edit a Doozy

Open a Doozy and choose Edit. Changes apply to future work and to new decisions made during ongoing work. A to-do already in progress may have acted on its earlier instructions, so send a direct message when a change must affect the current run.

Use the conversation history and to-do activity to validate important changes. If you tighten an approval boundary, review pending approvals before resuming work.

Attach role context

Attach documents that should remain available to the role, such as a style guide, escalation policy, architecture overview, account rubric, or reporting template.

Role files differ from files attached to one conversation: they’re durable context for that Doozy. Keep them current and remove superseded versions so the Doozy doesn’t have to guess which policy wins.

Duplicate a Doozy

Duplicate a Doozy when you need a similar role in the same workspace or want to test a material change without replacing the current role.

Duplication copies configuration such as the job description, skills, responsibilities, working hours, and toolkit selection. It doesn’t create a private Morph environment, copy every conversation, or turn workspace-shared credentials into isolated credentials.

Rename and review the duplicate before assigning work. Two indistinguishable Doozies with overlapping responsibilities make orchestration harder to understand.

Archive or delete a Doozy

Archive a Doozy when you may need its history or configuration later. Delete only when the role should be permanently removed.

Removing a Doozy stops its future responsibilities and schedules and removes it from active selection. It doesn’t automatically erase workspace files, to-do records, shared Morph state, or artifacts the Doozy created. Review those separately when removing access or sensitive data.

What a Doozy remembers

A Doozy’s context has several layers:

• its job description and attached role files;
• explicitly saved memories and user corrections;
• skills loaded for relevant work;
• active responsibilities and to-dos;
• conversation history and workspace records it can search;
• shared files and state in the workspace’s Morph environment.

Memories (aka saved preferences or long-term agent memory) are personal to the Doozy. A useful memory is concise and actionable, such as a repeated formatting preference or a correction that should affect future work. Other Doozies don’t automatically receive private memories, but they can receive the relevant facts through a to-do handoff or shared workspace artifact.

You can review, edit, refine, or delete saved memories from the Doozy’s memory view.

Organize a team of Doozies

Start with the smallest team that gives the work clear ownership. Split a role when its responsibilities require different evidence, permissions, or quality standards. Merge or archive roles when the same work is repeatedly duplicated.

For projects involving several Doozies, name one Doozy as the accountable coordinator. Other Doozies can investigate or produce components, but the coordinator should reconcile conflicts and deliver one coherent result.

Frequently Asked Questions